Graph Level Anomaly Detection

Graph-level anomaly detection aims to identify graphs that deviate significantly from a set of normal graphs, a crucial task with applications across diverse fields like fraud detection and medical diagnosis. Current research emphasizes improving detection accuracy by addressing class imbalance through techniques like counterfactual augmentation and by leveraging the spectral properties of graphs using Graph Neural Networks (GNNs) and other architectures such as transformers and normalizing flows. These advancements are improving the robustness and explainability of anomaly detection models, leading to more reliable and interpretable results in various domains.